<ul><li>Language models have emerged as powerful predictors of the viability of biological sequences.</li><li>In-context learning can distort the relationship between fitness and likelihood scores of sequences.</li><li>This distortion is prominently seen in sequences containing repeated motifs.</li><li>The phenomenon affects transformer-based models and is mediated by a look-up operation.</li></ul>

In-Context Learning can distort the relationship between sequence likelihoods and biological fitness

Discover more